Research map / Label-efficient and robust learning / Recognition and long-tail
Aligned Cross Entropy for Non-Autoregressive Machine Translation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace cross entropy with aligned cross entropy | CMLM · WMT'14 EN-DE test | BLEU: 18.05 -> 23.53 | Helped |
| Replace cross entropy with aligned cross entropy | CMLM · WMT'14 DE-EN test | BLEU: 21.83 -> 27.90 | Helped |
| Replace cross entropy with aligned cross entropy | CMLM · WMT'16 EN-RO test | BLEU: 27.32 -> 30.75 | Helped |
| Replace cross entropy with aligned cross entropy | CMLM · WMT'16 RO-EN test | BLEU: 28.20 -> 31.54 | Helped |
| Replace cross entropy with aligned cross entropy | CMLM · WMT'17 EN-ZH test | SacreBLEU: 24.23 -> 30.88 | Helped |
| Replace cross entropy with aligned cross entropy | CMLM · WMT'17 ZH-EN test | BLEU: 13.64 -> 19.79 | Helped |
| Use partially observed input instead of fully unobserved input during training (ablation) | AXE CMLM · WMT'14 EN-DE validation | BLEU: 21.97 -> 22.80 | Helped |
| Use partially observed input instead of fully unobserved input during training (ablation) | AXE CMLM · WMT'14 DE-EN validation | BLEU: 26.32 -> 27.59 | Helped |
| Calculate training loss on masked tokens only rather than all tokens (ablation) | AXE CMLM with partially observed input · WMT'14 EN-DE validation | BLEU: 22.80 -> 23.13 | Helped |
| Calculate training loss on masked tokens only rather than all tokens (ablation) | AXE CMLM with partially observed input · WMT'14 DE-EN validation | BLEU: 27.59 -> 28.01 | Helped |
| Increase skip-target penalty from δ=1 to δ=5 (ablation) | AXE CMLM · WMT'14 EN-DE validation | BLEU: 22.60 -> 23.13 | Helped |
| Increase skip-target penalty from δ=1 to δ=4 (ablation) | AXE CMLM · WMT'14 DE-EN validation | BLEU: 26.84 -> 28.01 | Helped |
Closest papers by meaning
- Order-Agnostic Cross Entropy for Non-Autoregressive Machine Translation (ICML 2021 · similarity 0.93)
- Non-Monotonic Latent Alignments for CTC-Based Non-Autoregressive Machine Translation (NeurIPS 2022 · similarity 0.90)
- AEQA-NAT : Adaptive End-to-end Quantization Alignment Training Framework for Non-autoregressive Machine Translation (ICML 2025 · similarity 0.88)
- Reconsidering Positional Supervision in Masked Diffusion Language Model Training (NeurIPS 2026 · similarity 0.87)
- Aligning Large Language Models with Representation Editing: A Control Perspective (NeurIPS 2024 · similarity 0.86)
- Decoding-Time Language Model Alignment with Multiple Objectives (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.